This paper describes a novel apparatus and method for conducting torsion tests on large-size hollow reinforced concrete (RC) beams. Previous torsion tests of concrete beams often used metal components to clamp the concrete beam specimen while exerting a twisting torque. This type of apparatus could allow gaps and slips between the clamping metal components and the concrete specimen, introducing errors into the small deformation measurement. In this study we used a monolithic transverse RC beam to convey a twisting torque without gaps to the twist end of the longitudinal test beam and constructed a monolithic RC block at the fixed end. Pre-tensioning forces were used to fasten the RC blocks of the specimens to the underlying steel seat and to the strong floor, thus producing a slippage-free mechanism for the fixed end. The test apparatus also incorporated devices to eliminate any constraints on the longitudinal elongation of the test beams. Seven quantities were experimentally determined for each of the specimens through continuous measurement, using various measurement lengths. The test results show that the test apparatus is rather efficient in determining small twists and that the initial branches of the obtained torque-twist curves exhibit a high degree of linearity. The results also demonstrate that a measurement length should be long enough to incorporate the entire failure region, but not too long, as too much length can cause the twist angles to be underestimated. Furthermore, the experimental values of the shear flow zone thickness decrease drastically around the cracking and remain relatively constant after cracking. The obtained elongation-twist curves are smooth when the measurement length covers the entire failure region, but otherwise they have abrupt turns due to drastic losses of measured twist angles.